AI researcher @kimmonismus posted that Anthropic's Opus 5 generated a complete playable game from a single prompt, consuming 690 million tokens at a $423 cost. The demonstration shows a workflow that previously required a development studio and a Steam launch.
Key facts
- 690 million tokens consumed in one prompt
- $423 total API cost per @kimmonismus
- Opus 5 model from Anthropic
- One person, few hours, small budget
- Previously required dev team and Steam release
Key Takeaways
- Opus 5 generated a playable game from one prompt, using 690M tokens at $423, per @kimmonismus.
- One person replaced a dev team.
What the $423 Prompt Actually Produced
The tweet According to @kimmonismus claims Opus 5 turned one prompt into a full game — 690 million tokens of compute, $423 in API spend, no human coding. The attached clip shows a playable title that, in the author's words, "would have had to be expensively developed and then sold on Steam in the past."
Token counts of that magnitude imply a long agentic loop — likely hundreds of model calls for code generation, asset creation, and iterative debugging — rather than a single completion. The $423 figure, at roughly $0.60 per million tokens for Opus-class output, suggests heavy use of extended thinking or multiple retries.
The result is less a technical benchmark than an economic one. The marginal cost of a polished interactive product has collapsed to the price of a mid-range dinner. For indie studios, the comparison is stark: a Steam-quality title typically costs tens of thousands in salaries and months of iteration.
Why This Changes the Unit Economics of Content
This is not the first time a frontier model has spat out a demo. What's notable is the token spend — 690 million is orders of magnitude above typical single-turn generations, implying the model autonomously drove a multi-hour build. The cost per token is now low enough that such loops are affordable for individuals.
The tweet does not disclose the game's genre, fidelity, or how much human steering occurred after the first prompt. Nor does it specify which Opus 5 variant (standard vs. extended thinking) was used. Those gaps matter: a curated, heavily re-prompted session would undercut the "one prompt" framing.
Still, the trajectory is clear. If a single agentic run can produce a saleable artifact for under $500, the barrier to entry for game development — and by extension many creative software products — drops to near zero. The question is whether the output quality holds up beyond a demo clip.
What to Watch
Watch for Anthropic's official Opus 5 pricing and context-window disclosure, expected in the coming weeks. If the $423 figure scales predictably, expect a wave of single-prompt game releases on itch.io and Steam — and a corresponding debate over what counts as "authorship."
[Updated 02 Aug via the_decoder]
The Decoder reports Opus 5 generated complete 3D games — a first-person shooter, a kart racer, and a Minecraft clone — from single prompts, with geometry, textures, physics, and music produced as code and run directly in the browser. In side-by-side comparisons, Opus 5 delivered significantly more detailed results than GPT-5.6 Sol and Kimi K3, all without external assets.
